InternAgentHarness: A Scalable Synthetic Environment for Enhancing LLM Agentic Abilities
About
Large language models (LLMs) are increasingly expected to act as generalist agents capable of solving complex real-world problems. Training such agents, however, requires stable and diverse environments that support repeated interaction with stateful, tool-augmented tasks and provide verifiable feedback. Despite recent progress, the development of robust LLM agents remains limited by the lack of realistic, scalable, and executable training environments. We present InternAgentHarness, a scalable synthetic environment for improving the agentic capabilities of LLMs. Built upon the InternBootcamp training framework~\citep{internbootcampv1}, InternAgentHarness instantiates executable agent environments through a four-layer interface that unifies prompt generation, tool execution, interaction control, and reward computation. We further introduce \bootcampcli, an agent harness that automatically converts diverse agentic tasks into a Bootcamp-trainable paradigm. Unlike static benchmarks, InternAgentHarnessmakes evaluation actionable: observed failures can be systematically converted into new synthetic tasks, filtered trajectories, reinforcement learning rollouts, and subsequent re-evaluation under the same executable interface. We instantiate InternAgentHarness on a suite of 10 tasks covering both text-only and vision-based agent scenarios. Starting from Qwen3-VL-30B-A3B-Thinking, both supervised fine-tuning (SFT) and reinforcement learning (RL) on InternAgentHarness substantially improve performance over untuned model. These results suggest that InternAgentHarness provides a practical foundation for scalable synthetic agent environments and enables the continuous improvement of LLM agents through an iterative cycle of evaluation, synthesis, training, and refinement.
Key insights
What problem does InternAgentHarness solve?
It provides a scalable, executable synthetic environment for training LLM agents. A four-layer interface unifies prompt generation, tool execution, interaction control, and reward computation, making evaluation failures reusable for training.
How does it differ from prior benchmarks?
Unlike static benchmarks, it converts failures into new tasks and training data. On a 10-task suite, SFT improved Qwen3-VL-30B's text-only average from 3.5 to 31.8; subsequent RL raised it to 51.4.
What are the practical uses of these findings?
The environment supports iterative agent improvement through SFT and RL. Feedback-rich tasks like battery and finance provided the strongest training signals, guiding future synthetic environment design for better closed-loop agent behavior.
What limitations do the authors acknowledge?
The current suite covers only 10 tasks focused on reasoning and visually grounded scenarios. Sparse-feedback tasks remain challenging, and extending to embodied agents, multi-agent collaboration, and human-in-the-loop refinement is needed.